The population pharmacokinetics of lithium was determined. Psychiatric inpatients who were at least 18 years old, had normal renal function, and were receiving lithium two or three times daily were the subjects of the study; 79 patients met the criteria. Serum lithium concentrations were measured by flame photometry 12 hours after the evening dose. Other data collected included age, gender, race, height, weight, serum creatinine and sodium concentrations, diet, and concurrent drug use. The nonlinear mixed-effects model (NONMEM) program was used to estimate lithium clearance, lithium volume of distribution, and error variability and to determine the influence of patient characteristics on lithium clearance. With the initial model, the mean lithium volume of distribution was 32.8 L and mean lithium clearance was 1.36 L/hr. With an intermediate model, lithium clearance estimates improved on the basis of patient size (weight and body surface area), daily lithium dosage, age, gender, and race. When only the most significant variables--lean body weight and creatinine clearance--were retained, a final model was obtained that yielded a coefficient of variation for lithium clearance of about 24% and gave fairly accurate predictions of steady-state lithium concentrations (coefficient of variation, about 16%). Analysis of lithium pharmacokinetics with the NONMEM program suggested that lean body weight and creatinine clearance are important predictors of lithium clearance.